SAR image matching based on rotation-invariant description
Yunhao Chang, Qing Xu, Xin Xiong, Guowang Jin, Huitai Hou, Dan Man
Abstract
The utilization of scale invariant feature transform algorithm in synthetic-aperture radar images (SAR-SIFT) to match image features may lead to principal orientation assignments of descriptors being affected by speckle noise, thereby diminishing accuracy. In this study, we propose using the Fourier histogram of oriented ratio gradient (Fourier HORG) descriptor for robust matching of SAR images. This method is based on the SAR-SIFT algorithm framework. During feature description, the rotation-invariant Fourier HORG descriptor is established by performing Fourier analysis on the ratio gradient in the polar coordinate system, whereby the principal orientation assignment process is avoided and the robustness of SAR image registration improved. A matching experiment was conducted involving four sets of SAR image pairs, and the results demonstrated that our method exhibited higher accuracy and robustness compared to image matching based on the Fourier histogram of oriented gradient (Fourier HOG) descriptor and the SAR-SIFT algorithm, thus confirming the effectiveness of our proposed method.